Postediting

Page 1

More About Post-Editing By Mike Dillinger

Machine

translation (MT) software is here to stay. So what is a translator to do? Not a lot, really. If you do very specialized marketing or literary translations, you probably will not see much MT software being used. But if you work with product or training information for global companies, for example, my crystal ball says that you will hear about MT or post-editing sometime soon. My colleague Laurie Gerber did a great job of summarizing lots of information about MT during her ATA webinar this past April and in an article in the November/December 2011 issue of this magazine.1 I would like to add some more information to the discussion, based on recent post-editing workshops I gave for the Northern California Translators Association, at ATA’s 52nd Annual Conference in Boston, and for other groups.

Translation Memory and Machine Translation To work comfortably with MT software, it helps when you understand what it really does. Machine translation software is just like the translation memory (TM) products with which we are already familiar. It does not really translate; it only helps us reuse

20

To work comfortably with machine translation software, it helps when you understand what it really does. words and segments that have already been translated. That’s it. There is no android exterminator hiding in the software, no black magic, no cybermega-brain. Here is a breakdown of the main capabilities of MT. Good Matches: The TM products that are so familiar to us check new sentences against the existing segments that we put in memory and then return the good matches. We use our skills to double-check whether the “good” matches make sense in the context or need tweaking and polishing. Machine translation software does the same kind of matching. In fact, with products like those from Systran and ProMT, you can actually plug the TMs that you already have into your MT system. The system will check for good matches in your memories before it does anything else.

Fuzzy Matches: If there is no good match, we can have recent TM products “assemble” a translation from text fragments in order to give us more suggestions from which to work. We use these half-baked suggestions to craft good translations faster, even though we end up ignoring a lot of what is suggested. Machine translation software does the same thing: it assembles translations from bits and pieces, but in a much more sophisticated way. Machine translation software extends TM technology with the ability to build more and better “assembled” translations. So we get more and better suggestions with which to work. Non-Matches: Finally, if a TM product cannot find enough information to even guess at a translation, it gives up and does not suggest anything. Machine translation software is

The ATA Chronicle

n

January 2012


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.